feat(gtm): v2.5 - Hard Fact Extraction & UI

- Backend: Implemented secondary extraction phase for structured specs (JSON schema).

- Backend: Added strict normalization rules (min, cm, kg).

- Frontend: Added 'Phase1Data' interface update for specs.

- Frontend: Implemented new UI component for 'Technical Specifications' in Phase 1.

- Frontend: Updated header and sidebar to display 'v2.5' build marker.

- Docs: Updated architectural documentation.
This commit is contained in:
2026-01-06 19:36:42 +00:00
parent 0995d7c43a
commit 02b02856f6
6 changed files with 268 additions and 4 deletions

View File

@@ -203,6 +203,90 @@ def phase1(payload):
try:
data = json.loads(response)
# --- PART 2: HARD FACTS EXTRACTION ---
spec_schema = """
{
"metadata": {
"product_id": "string (slug)",
"brand": "string",
"model_name": "string",
"category": "cleaning | service | security | industrial",
"manufacturer_url": "string"
},
"core_specs": {
"battery_runtime_min": "integer (standardized to minutes)",
"charge_time_min": "integer (standardized to minutes)",
"weight_kg": "float",
"dimensions_cm": { "l": "float", "w": "float", "h": "float" },
"max_slope_deg": "float",
"ip_rating": "string",
"climb_height_cm": "float",
"navigation_type": "string (e.g. SLAM, LiDAR, VSLAM)",
"connectivity": ["string"]
},
"layers": {
"cleaning": {
"fresh_water_l": "float",
"dirty_water_l": "float",
"area_performance_sqm_h": "float",
"mop_pressure_kg": "float"
},
"service": {
"max_payload_kg": "float",
"number_of_trays": "integer",
"display_size_inch": "float",
"ads_capable": "boolean"
},
"security": {
"camera_types": ["string"],
"night_vision": "boolean",
"gas_detection": ["string"],
"at_interface": "boolean"
}
},
"extended_features": [
{ "feature": "string", "value": "string", "unit": "string" }
]
}
"""
specs_prompt = f"""
PHASE 1 (Part 2): HARD FACT EXTRACTION
Input: "{analysis_content}"
Task: Extract technical specifications strictly according to the provided JSON schema.
NORMALIZATION RULES (STRICTLY FOLLOW):
1. Time: Convert ALL time values (runtime, charging) to MINUTES (Integer). Example: "1:30 h" -> 90, "2 hours" -> 120.
2. Dimensions/Weight: All lengths in CM, weights in KG.
3. Performance: Area performance always in m²/h.
4. Booleans: Use true/false (not strings).
5. Unknowns: If a value is not in the text, set it to null. DO NOT HALLUCINATE.
LOGIC FOR LAYERS:
- If product uses water/brushes -> Fill 'layers.cleaning'.
- If product delivers items/trays -> Fill 'layers.service'.
- If product patrols/detects -> Fill 'layers.security'.
EXTENDED FEATURES:
- Put any technical feature that doesn't fit the schema into 'extended_features'.
Output JSON format ONLY based on this schema:
{spec_schema}
"""
log_and_save(project_id, "phase1_specs", "prompt", specs_prompt)
specs_response = call_gemini_flash(specs_prompt, system_instruction=sys_instr, json_mode=True)
log_and_save(project_id, "phase1_specs", "response", specs_response)
try:
specs_data = json.loads(specs_response)
data['specs'] = specs_data
except json.JSONDecodeError:
logging.error(f"Failed to decode JSON from Gemini response in phase1 (specs): {specs_response}")
data['specs'] = {"error": "Failed to extract specs", "raw": specs_response}
db_manager.save_gtm_result(project_id, 'phase1_result', json.dumps(data))
# WICHTIG: ID zurückgeben, damit Frontend sie speichert